Design a CNN-GRU Model for Handwritten Character Recognition and Performance Evaluation on Char74k Dataset

In recent times, new methods for character recognition have been developed as a result of progress in artificial intelligence and computer vision. Even though humans are able to recognize characters, objects, and people with high precision, machines face challenges in this area, particularly when dealing with the different patterns and shapes of characters across different languages. Researchers have used various techniques to assist machines in character recognition, including analyzing input images, identifying features, implementing classification, and training neural networks. This research paper suggests a novel hybrid approach to character recognition that outperforms previous methods. The hybrid model combines a CNN with GRU to improve accuracy and speed. The GRU substitutes the fully connected layer of the CNN, dynamically classifying the characteristics extracted from the input image. The model is assessed using the Char74k dataset, achieving a training accuracy of 99.28% and a validation accuracy of 96.43%.